Study On Air Quality Forecast Model In Typical City Of Southern And Northern China

In recent years, there is a wide range of heavy pollution weather for a long timein many parts of China, which occurred mainly in Pearl River Delta,Beijing-Tianjing-Hebei region and Yangtze River Delta. Therefore it is important todevelop a real-time and high-effective air quality forecast model, which not only canguide residents to travel, but also guide the functional department to prevent andcontrol the heavy pollution weather and offer the relative technical support. Thoughthe air quality forecast model is one of the hotspots in current study, traditionalstatistic forecast model and foreign numerical forecast model are not applicable to thecurrent situation of heavy regional air pollution status in China, due to the reason oftechnical flaws and systematic obstacles. Aiming at Southern and Northern China, anair quality forecast model is developed based on the dynamic sample selectionmechanism and BP neural network in this thesis, and about the method of the spatialvisualization is also analyzed.Firstly, three important cities from Pearl River Delta and Beijing-Tianjing-Hebeiregion repectively are chosen separately. The characteristic of temporal distribution,the autocorrelation and the correlation with the meteorological factors of the regularair pollutants, including SO₂, NO₂, PM₁₀, CO, O₃, and PM_2.5, are analyzed. Resultsshow that the air pollution is more serious in Beijing-Tianjing-Hebei region citiesthan in Pearl River Delta cities, and the proportion of particulate matter pollution inBeijing-Tianjing-Hebei region cities is greater, and the proportion of ozone pollutionin Pearl River Delta cities is greater as a whole.Then, based on the analysis results of the impact factors of different cities anddifferent pollutants above, the related parameters of the dynamic sample selectionmechanism and BP neural network are determined. Air quality forecast model for aweek’s results of No.5Middle School monitoring station in Guangzhou and Tiantan monitoring station in Beijing is proposed. The model is used in the two sites aboutdifferent pollutants and the forecasting results of three days after were analyzed. Theresults show that the mean relative error of Shiwuzhong monitoring station rangefrom0.18to0.34, which the overall forecasting results are better. The forecastingresults of CO and NO₂are better than other. To the CO and NO₂: the mean relativeerror of the forecasting results are more than0.31, and the forecast grade accuracy ismore than92%; To the PM_2.5and O₃: the mean relative error of the forecasting resultsranges from0.21to0.47, and the forecast grade accuracy is from70%to90%. Theforecasting accuracy of the model proposed is higher than that of the multiple linearstepwise regression model.Last, the method of spatial visualization is made based on ArcGIS. This is to say,comparative analysis of three interpolation techniques is made in Pearl River Delta,including two deterministic and one geo-statistical method, inverse DistanceWeighted （IDW）, and Completely Regularized Spline （CRS） and Kriging （OK）respectively. Combined with the cross-validation method, the three methods arecompared to analyze the accuracy of interpolation and intuitive effect of spatialdistribution from some aspects, including properties of air pollutants, characteristicsof seasons and so on. The results show that Kriging always could draw an optimalresult from the point of accuracy, but the smoothness of the transition area in spatialdistribution of air quality is poor. The method is more applicable for the work of theregional air quality forecast. The properties of air pollutants and characteristics ofseasons have a significant impact on the interpolation results. The interpolationresult of secondary pollutant is significantly better than of primary pollutant. Themean relative error of the interpolation ranges from0.186to0.313, and thecorrelation coefficient is above0.8at most time.